主动学习的科技文献研究对象标引体系研究

Objective] This study aims to identify the research object attribute instance from the paper titles. With the help of limited labeled samples, we could maximumize the accuracy of research object recognition. [Methods] We first analyzed the grammatical features of scientific research objects based on conditional random field sequence labeling algorithm. Second, we recognized and extracted research objects using a small amount of samples. Finally, we introduced an active learning iterative labeling system based on unlabeled data to improve the research object recognition accuracy. [Results] The results showed that the proposed method could efficiently use the unlabeled data, and increase the accuracy of the research object recognition to 78.3%. [Limitations] The proposed algorithm needs to be further optimized to improve its efficiency. [Conclusions] The proposed method performed well on the research object attributes identification, which is the foundation for further mining the knowledge system and the structure of science and technology literature.